Image-Based Lateral Position, Steering Behavior Estimation, and Road Curvature Prediction for Motorcycles

Abstract

This paper presents an image-based approach to simultaneously estimate the lateral position of a powered-two wheeled vehicle on the road, its steering behavior and predict the road curvature ahead of the motorcycle. This work is based on the inverse perspective mapping technique combined with a road lanes detection algorithm capable of detecting straight and curved lanes. Then, a clothoid model is used to extract pertinent information from the detected road markers. Finally, the performance of the proposed approach is illustrated through simulations carried out with the well-known motorcycle simulator "BikeSim". The results are very promising since the algorithm is capable of estimating, in real time, the road geometry and the vehicle location with a better accuracy than the one given by the commercial GPS.

Learning a Curve Guardian for Motorcycles

“…For motorcycles specifically there has been less activity but nonetheless the field is benefiting from the technological improvements. [5] were the first to demonstrate image-based lateral position estimation within a lane, however they rely on traditional, geometric computer vision techniques. We believe formulating the task as a learning problem and utilizing the advances in CNNs will create a perception system that is able to generalize better for a diverse set of scenarios.…”

Vision-Based Lane Crossing Point Tracking for Motorcycles

“…The contribution of this paper is based on the results initially introduced in [14]. The reader could refer to the following videos https: //www.youtube.com/playlist?list= PLRTI62SuvNymK2Dx-YKha-1a4Sp54IVs8 for visual illustrations.…”

“…The reader could refer to the following videos https: //www.youtube.com/playlist?list= PLRTI62SuvNymK2Dx-YKha-1a4Sp54IVs8 for visual illustrations. In [14], the authors used the IPM technique combined with a road lanes filter allowing to generate a bird-eye-view of the road markers (as presented in figure 3). Then, a clothoid model of the road is used to extract pertinent information such that the P2WV relative lateral displacement and heading angle to the road markers.…”

“…Both parametres allow to faithfully reconstruct the road trajectory in the selected Region Of Interest (ROI). Note that, even if the ROI limit ahead of the P2WV is chosen about 30 meters (see [14]), each road marker trajectory can be extended since we know its third degree polynomial approximation.…”

“…Whereas, in the simulations discussed in [14], the right road marker is defined as a static reference, we proposed to introduce a dynamic reference. Indeed, the accuracy of the lane i trajectory reconstruction mainly depends on two factors: the proximity with this lane and its attribute (dashed or solid).…”